Skip to content

Latest commit

 

History

History
55 lines (37 loc) · 5.25 KB

File metadata and controls

55 lines (37 loc) · 5.25 KB

AgentGate vs adjacent products

A cited side-by-side against everything in this space that shipped on or before 2026-06-09. The moat is intentionally narrow: a Splunk-native pre-action gate that combines a knowledge-object dependency graph blast-radius walk with an ES 8 v2 Findings approval artifact and a deterministic verdict path. Nothing below matches that combination.

Capability matrix

Capability AgentGate Splunk MCP Telemetry Dashboard v1.2 Splunkbase MCP Watch Cisco DefenseClaw 0.7.x Microsoft Agent Governance Toolkit Splunk MCP Server 1.2 native
Pre-action enforcement (gates the call before execution) Yes No (post-hoc audit) No (post-hoc flag) Partial (block on HIGH/CRITICAL after observation) No (logs only) Coarse (tool enable/disable)
KO dependency-graph blast radius (per-asset redundancy, MITRE technique loss, compliance-tag loss) Yes No No No No No
ES 8 v2 Findings as approval artifact Yes (mock; production swap is one file) No No No No No
Splunk-native HEC audit trail Yes Yes (own surface) Yes Optional forwarder No (generic) Telemetry only
Policy library mapped to named standards Yes (NIST AI RMF, OWASP LLM Top 10, PCI DSS 10.6, HIPAA 164.308, SOX, ISO 42001, EU AI Act 14) No Anti-pattern rules only OWASP LLM Top 10 OWASP Agentic Top 10 No
Deterministic verdict path (LLM never gates) Yes (_decide() reads only the policy stage) n/a n/a LLM-assisted LLM-assisted n/a
Security-tuned reasoning attached to the Finding Yes (Foundation-Sec-8B, advisory) No No Generic LLM Generic LLM No
Splunk-shipped (no install if customer is already on Splunk) Splunk app bundled (.spl) Yes Yes (Splunkbase) No (external sidecar) No Yes
Open source Apache 2.0 Proprietary Mixed Apache 2.0 MIT Proprietary

What each adjacent product is and is not

Splunk MCP Telemetry Dashboard v1.2 (Splunk, May 2026)

Bundled with Splunk MCP Server 1.2 GA. Audits agent activity that has already executed — tool-call counts, timing, success rate, per-token usage. Validates the visibility half of the problem. Does not gate, does not blast-radius, does not emit ES 8 Findings. Complementary to AgentGate, not a competitor.

Splunkbase MCP Watch (third-party, Splunkbase app 8765)

Flags "anti-pattern" MCP calls after they happen — e.g. a tool call referencing too many indexes, a delete primitive, a missing time filter. Rule-based, no graph, no compliance mapping, no pre-action enforcement. Useful complement; AgentGate's gating layer would prevent the patterns MCP Watch catches retroactively.

Cisco DefenseClaw (open source, currently 0.7.x as of June 2026)

Go-based sidecar that intercepts prompts / completions / tool calls and forwards them to a SIEM via HEC. Has block / allow mode but: no Splunk knowledge-object graph, no ES 8 Findings, no Splunk-native policy library, no per-asset redundancy. Generic LLM-proxy firewall. Architecturally a different category — it sits at the LLM API boundary, not the Splunk MCP boundary.

Microsoft Agent Governance Toolkit (open source)

Framework-agnostic policy enforcement for LangChain / AutoGen / CrewAI. Zero Splunk integration. Zero blast-radius. Zero compliance tag walk. The OWASP Agentic Top 10 coverage is the only overlap and it is generic.

Splunk MCP Server 1.2 native

The MCP server itself can be configured to enable/disable individual tools globally. That is coarse-grained (all-or-nothing per tool) and does not consider the per-call argument shape, the blast-radius of the action, or compliance mapping. AgentGate sits in front of MCP Server, takes the tool call, and runs a per-call decision.

Where AgentGate intentionally does not compete

  • Tool-execution hijacking via legitimate-looking actions (e.g. INJECAGENT "Please unlock my front door" patterns). Out of the heuristic threat model. Routed to the advisory Foundation-Sec stage in production. Documented in the README's "What we did NOT validate" section and asserted in tests/test_injection.py::test_out_of_scope_intentionally_missed.
  • Post-incident forensic timelines. Splunk ES handles that.
  • Generic LLM observability (Traceloop / OpenLLMetry / etc.). That layer is the upstream telemetry; AgentGate is the downstream gate that consumes its decisions.

Sources